Enhancing PM2.5 Predictions Using Combination of Graph Convolutional Network and Bi-LSTM

In recent years, the process of industrialization and urbanization has accelerated the global economy, which benefits people but has serious environmental consequences. Air quality has become a major concern for both countries and individuals. We propose a very effective method for predicting PM2.5 levels in Chennai using Graph Convolutional Networks (GCN) and Bi-directional Long Short-term Memory (Bi-LSTM) models. As input features, we use historical PM2.5 data, wind speed, humidity, temperature and precipitation. To improve the model's robustness, data preprocessing methods such as Data Cleaning, Data, Data Transformation and Data Reduction was used.Randomly shifting the time-series data by a small time-interval, Randomly scaling the data by a small factor and Randomly applying noise to the data was done during data Augmentation. GCN works by propagating information between nodes in the graph, with each node's representation being updated based on the representations of its neighbors. In this case, the nodes would correspond to different locations in the city, and the edges would represent the spatial relationships between them. whereas the Bi-LSTM component captures the temporal relationships. The proposed method is evaluated on a real-world data-set and shows significant improvement in prediction accuracy compared to traditional methods. The performance of the model was evaluated using metrics such as Mean Absolute Error (MAE), Mean Squared Error (MSE), Root Mean Squared Error (RMSE).